A Cooperative HIde and SeeK DiscoverY over In Network Management Lucas Guardalben, Susana Sargento and Paulo Salvador Vitor Mirones Universidade de Aveiro/Instituto de Telecomunicac¸oes˜ Portugal Telecom Inovac¸ao˜ S/A Aveiro, Portugal Aveiro, Portugal guardalben,susana,[email protected] [email protected] Abstract—Future networks are expected to become more management generally uses dedicated servers to take decisions complex and dynamic than today’s networks. In these scenarios, and actions. a centralized approach for network management is extremely The vision of Autonomic Network Management [1] is being difficult and contains serious scalability problems. An alternative to solve these problems is to introduce the concept of In Network pursued by researchers as an answer to the previous problems. Management (INM), where each INM entity, a network node, has This is the capability of network entities to self-govern their the autonomy to self-govern its behavior. In this concept, each behavior within the constraints of business goals that the INM entity participates in a distributed management process, network, as a whole, seeks to achieve. To address all these which requires cooperation between nodes to monitor, analyze, concepts, new Internet architectures need to be designed, decide and act upon the network. However, to ensure the communication of nodes in In Network Management process, following the GENI initiatives in the US of Clean Slate several phases need to be performed, such as, discovery and boot- design approaches to re-build the Internet. In this sense, a new strapping, exchange of network information, dissemination of paradigm to ensure autonomic network behavior is In Network local decisions and final decision dissemination for enforcement. Management (INM)[2], [3], [4] that is based on a distributed In this paper we address the discovery and bootstrapping, and control approach. This paradigm was studied in the scope of exchange of information between nodes, towards the complete INM process. We propose Hide and Seek (H&S), a new algo- 4WARD project [5], [6] and describes initiatives that facilitate rithm for network discovery, and information propagation and the embedding of distributed management functionalities over synchronization. The results show that H&S spends fewer cycles INM entities. As opposed to traditional management, in INM to discover all nodes and records lower messages overhead for each entity interacts with its peers with the ability to take information synchronization when compared to non-controlled decisions based on the knowledge from the other elements [2] and probabilistic (gossip) flooding dissemination. (Fig. 1 ). I. INTRODUCTION In the future, different networks and technologies will interact in a global network of networks. Multiple access technologies will be simultaneously available. The choice of access network could depend on technical (achievable bitrate, maximum delay) or non-technical (cost) decision criterions. Network elements and terminals will vary from the sim- plest, e.g. a sensor, to the most complex, e.g. a server or a mobile router. The role and capabilities of such devices in the management overlay will depend on their characteristics. Cooperation and delegation between nodes will be required. Changes in the network can occur naturally when a node arrives or departures from the network, or can be caused by Fig. 1. Traditional (left side) versus In Network Management (right side) a malfunction somewhere in the network. There are similar- [4] ities in the handling of both processes. The ability to adapt to changes is nowadays performed by a traditional Internet Consequently, for the nodes to know each other and to management, through centralized approaches. However, they communicate information and management decisions in an should be provided without the need for human intervention. efficient way, some phases are required to provide the means The traditional Internet management approaches use exter- for the nodes to cooperate. Some of the phases comprise nal management that resides outside the network on servers discovery of nodes and bootstrapping of the network, exchange and stations that interact using standard protocols, such as of network information between nodes, dissemination of local SNMP (Simple Network Management Protocol) or CLI (Com- decisions from the nodes to other nodes, and final decision mand Line Interfaces). As depicted in Fig. 1 , the traditional enforcement dissemination to perform distributed enforcement of the cooperative decisions (section II). After completing all • A resource discovery algorithm for efficient dissemina- these phases, it is possible, in terms of communication, to tion of local connectivity information; embed INM concept over the network through INM architec- • Synchronization protocol that guarantee the connectivity tural entities. For the discovery, bootstrapping and exchange from local interactions; of network information, we propose the Hide and Seek (H&S) The major challenge in the design of a bootstrapping mech- approach, a new algorithm for network discovery, bootstrap- anism is to ensure scalability and robustness, which becomes ping, and information propagation and synchronization. H&S more complex and less efficient with the increasing number of is essential to ensure the communication to propagate the INM entities entering and exiting the network. It is necessary relevant and sufficient information between INM entities, to develop an efficient bootstrapping mechanism that contains necessary for the decision processes. these two requirements to be applied in the INM concept. The rest of this paper is structured as follows. Section The second phase requires the dissemination of informa- II summarises the In Network Management concept. Subse- tion between the INM entities to perform the management quently, the related work is described in Section III. Section decisions. The dissemination of information in the network IV presents the hide and seek algorithm. Next, the simulation also contains a widely set of approaches in the literature. setup is depicted in Section V and results are discussed in Some examples can use simple flooding [8], probability-based Section VI. Finally, Section VII summarises the important flooding [9], a minimum connected dominating set (MCDS) conclusions and future work. Based [10], location-based [11], epidemic-based [12] and II. IN NETWORK MANAGEMENT (INM) cluster-based [13]. Other types of approaches are necessary to solve constraints in flooding mechanisms [14]. In our The INM paradigm is based on a distributed control ap- approach, it is required an efficient mechanism that distributes proach. This approach provides continuous interactivity be- information between the nodes, considering the scope and type tween nodes in order to exchange information about each node of information. A mechanism is also required to distribute and its traffic (and therefore the network). This information local decisions to sets of nodes cooperating in the management will allow the network to make automatic decisions, reacting process and to disseminate the results of the decision process. to network changes (such as link failures, load variations) Beyond the communication process that defines to which INM and continuously optimizing the network resources (in both entity specific information should be sent, it is very important physical and virtual networks, to users and services) according to ensure the knowledge of information management, that is, to optimization mechanisms. Furthermore, to ensure nodes which set of information should be distributed to which nodes. knowledge and communication in INM process, four phases In the third phase, it is required to define the communication are envisaged, such as the discovery and bootstrapping, ex- process to disseminate the local management decisions to change of information, dissemination of local management provide global cooperative decisions between the INM entities. decisions and final management decision dissemination for Afterwards, it will be created primitives towards the optimized enforcement. communication process between the INM entities. In the first phase, the INM entities need to communicate to The last phase provides the dissemination of the final discover the network nodes and information, their roles, and decision that should be sent to the nodes in order to enforce how they should operate. As will be referred in Section III, this it. It is required to define also which nodes need to receive communication process should have low cost in a distributed the information to provide the required action, which nodes way. In this sense, we consider the bootstrapping as the initial need to access the information, and how to identify them to warm-up of the network (or a new INM entity) where each optimize the dissemination process. node makes the initial contact with its INM neighbours. Note that the discovery also refers to the continuous process of III. RELATED WORK:DISCOVERY AND BOOTSTRAPPING maintaining the information updated (including the network MECHANISMS status). The discovery of nodes and the network topology This section presents some relevant papers that discuss the are issues already addressed in the literature. In most of subjects of discovery and bootstrapping in different types of the the proposed mechanisms, the nodes send broadcasting networks. messages to all neighbor nodes to obtain information from the topology of the network. This process must be done